Research On Personalized Recommendation Algorithm Based On Behavior Characteristics Analysis

With the rapid development of Internet technology,data overload has become a major pain point in network life.How to extract high-value information for users from huge data has become an important direction of current information system research.Recommendation system,with its unique personalized services,has become an increasingly important application that people rely on in their online life.However,when implementing recommendations to users,problems such as cold start and interest drift have become difficult challenges in this field.The user’s behavior data in the recommendation system contains many hidden behavior characteristics,and the recommendation algorithm can be greatly improved by exploring these behavior characteristics.Firstly,the recommendation algorithm based on static behavior feature analysis is studied.Secondly,through the analysis of user memory characteristics,item popularity characteristics and interactive familiarity characteristics,a recommendation algorithm based on dynamic behavior characteristics analysis is proposed.Finally,a recommendation algorithm based on behavioral feature analysis is designed and implemented by combining the above two algorithms.The main contents of this paper are as follows:(1)In view of the failure to use the recommendation algorithm for calculation caused by the cold start problem,relevant objects cannot be recommended or the recommendation accuracy is low.This paper designs a recommendation algorithm based on static behavior feature analysis.By analyzing the user individual characteristic and the project tag information,establish a cold start users and individual characteristics,the connection between the project tag,through the K-means clustering algorithm to build similar set of users,the user focused by similarity measure to find the most similar users,finally by improving the prediction scoring method for cold start object is recommended.(2)By analyzing the current recommended interest drift problems existing in the system,the laws of human cognitive memory and ebbinghaus forgetting,give full consideration to the user of the project features and forgotten memories,the user interest psychological modeling calculation according to the cognitive memory and forgetting,finally build the user memory feature model.Considering the contextual and temporal factors when the project is recommended,the model of project popularity characteristics is constructed by imitating the user’s cognitive psychology of seeking popularity.According to the present good and bad are intermingled of users on the platform of the project evaluation and evaluation of the phenomenon of distortion,analyze the behavioral data,on the basis of interaction data and attribute information of the project,the field of building user familiar with the features and project attributes,and familiar with the modeling calculation to build interactive features.Finally,user memory features,item popularity features,interactive familiarity features and collaborative filtering algorithm are integrated into a recommendation algorithm based on dynamic behavior feature analysis.(3)Finally,the algorithm proposed above is organically combined into a recommendation algorithm based on behavioral feature analysis.The algorithm proposed in this paper is compared with the traditional IB-CF and PMF algorithms in the relevant indexes.Experiments show that the recommendation algorithms proposed in this paper have a certain improvement in the accuracy of recommendation compared with them.